Adaptive real-time state space control of a neurostimulation device

Abstract

Neurostimulation, such as electrical and/or magnetic neurostimulation, is controlled using an adaptive real-time state space (ARTISTS) control framework to determine and/or adjust stimulation settings (e.g., stimulation waveforms). The ARTISTS control framework generally includes an adaptive autoregressive model of the nervous system's response to stimulation, an amended Kalman filter to estimate the state and coefficients of the autoregressive model, and a linear quadratic regulator to determine the stimulation waveform to be delivered.

1 . A method for determining stimulation settings for a neurostimulation device, the method comprising:
 (a) measuring neural activity in a subject using a recording electrode, the neural activity being in response to a neurostimulation delivered to a neural target in the subject by a neurostimulation device;   (b) constructing an autoregressive model based on the measured neural activity;   (c) estimating a neural state of the subject based on the autoregressive model, wherein the neural state comprises a real-time state of the subject's nervous system;   (d) determining stimulation settings based on the estimated neural state; and   (e) delivering neurostimulation to the neural target in the subject with the neurostimulation device using the stimulation settings.   
     
     
         2 . The method of  claim 1 , wherein the autoregressive model is constructed based on the measured neural activity to minimize prediction error of a response to the neurostimulation delivered to the neural target. 
     
     
         3 . The method of  claim 1 , wherein the neural state is estimated using a Kalman filter constructed based on the autoregressive model. 
     
     
         4 . The method of  claim 3 , further comprising extracting model coefficients from the autoregressive model and constructing the Kalman filter using the model coefficients. 
     
     
         5 . The method of  claim 4 , wherein the Kalman filter is constructed using the model coefficients to generate Kalman filter coefficient matrices and Kalman gain for the Kalman filter. 
     
     
         6 . The method of  claim 1 , wherein the stimulation settings are generated using a linear quadratic regulator. 
     
     
         7 . The method of  claim 6 , wherein the linear quadratic regulator is constructed using the estimated neural state to define a cost function of the linear quadratic regulator. 
     
     
         8 . The method of  claim 7 , wherein the linear quadratic regulator is further constructed based on model coefficients extracted from the autoregressive model. 
     
     
         9 . The method of  claim 1 , wherein the neural activity is measured using one or more recording electrodes. 
     
     
         10 . The method of  claim 9 , wherein the one or more recording electrodes are part of the neurostimulation device. 
     
     
         11 . The method of  claim 1 , wherein the stimulation settings comprise a neurostimulation waveform. 
     
     
         12 . A controller for controlling a neurostimulation device, comprising:
 an input that receives neural activity signals measured from a subject;   a processor to:
 receive the neural activity signals from the input; 
 construct an autoregressive model based on the neural activity signals; 
 extract coefficients from the autoregressive model; 
 estimate a neural state based on the extracted autoregressive model coefficients, wherein the neural state comprises a real-time state of the subject's nervous system; 
 determine stimulation settings based on the estimated neural state; and 
   an output that receives the stimulation settings from the processor and communicates the stimulation settings to a neurostimulation device.   
     
     
         13 . The controller of  claim 12 , wherein the autoregressive model is constructed by the processor to minimize prediction error of a response to neurostimulation delivered to a neural target in a subject. 
     
     
         14 . The controller of  claim 12 , wherein the neural state is estimated using a Kalman filter constructed based on the extracted autoregressive model coefficients. 
     
     
         15 . The controller of  claim 14 , wherein the processor constructs the Kalman filter by using the model coefficients to generate Kalman filter coefficient matrices and Kalman gain for the Kalman filter. 
     
     
         16 . The controller of  claim 12 , wherein the processor generates the stimulation settings using a linear quadratic regulator. 
     
     
         17 . The controller of  claim 16 , wherein the linear quadratic regulator is constructed by the processor using the estimated neural state to define a cost function of the linear quadratic regulator. 
     
     
         18 . The controller of  claim 17 , wherein the linear quadratic regulator is further constructed by the processor based on the extracted autoregressive model coefficients. 
     
     
         19 . The controller of  claim 12 , wherein the neural activity signals are received by the input from one or more recording electrodes. 
     
     
         20 . The controller of  claim 12 , wherein the stimulation settings comprise a neurostimulation waveform.